This repository includes some basic Torch code for evaluation and visualization of our network output. A pretrained model is available on the project site at www-personal.umich.edu/~alnewell/pose. Include it in the main directory of this repository to run the demo code.
In addition, if you download the full MPII Human Pose dataset and replace images/ you can generate full predictions on the validation and test sets.
To run this code you must have installed:
- Torch7
- hdf5 (and the torch-hdf5 package)
- cudnn
- qlua (for displaying results)
(If there are any other dependencies that I have (likely) forgotten, get in touch and I will list them here)
For displaying the demo images:
qlua main.lua demo
For generating predictions:
th main.lua predict-[valid or test]
For evaluation on a set of validation predictions:
th main.lua eval